Photodiode Wizard and the SPICE model it generates give very different noise levels

I realize the photodiode wizard is approximate while SPICE is a little more rigorous, but the difference for this circuit is very large:

A low level of noise is estimated:

If I download the spice file for that circuit and run it in ltspice, then integrate noise over the target bandwidth, I get:

SPICE predicts 3x worse noise, and furthermore, SPICE is assuming a brickwall filter at 30 MHz, so real performance would be worse.  Am I doing something wrong in SPICE or is the photodiode wizard just really inaccurate?